Skip to content

Transforms complex documents like PDFs and Office docs into LLM-ready markdown/JSON for your Agentic workflows.

test(pdf): cover searchable scan and digital background auto routing

myhloli680ff2aec537
diff --git a/tests/unittest/test_pdf_scan_text_routing.py b/tests/unittest/test_pdf_scan_text_routing.py
new file mode 100644
index 00000000..59d36848
--- /dev/null
+++ b/tests/unittest/test_pdf_scan_text_routing.py
@@ -0,0 +1,69 @@
+"""通过真正的 DocVortex 分类器核验 MinerU auto 对扫描补文本与数字背景页的路由。"""
+
+from io import BytesIO
+from types import SimpleNamespace
+
+import pytest
+from PIL import Image
+from reportlab.lib.utils import ImageReader
+from reportlab.pdfgen.canvas import Canvas
+
+from mineru.backend.analysis.pdf import pipeline
+from mineru.model.runtime import hybrid as hybrid_runtime
+
+
+def _image_layer_pdf(kind: str) -> bytes:
+ """独立构造背景正文、隐藏搜索层、透明文字和被图像遮住的文字,避免引用另一个仓库的测试。"""
+ stream = BytesIO()
+ painter = Canvas(stream, pagesize=(400, 300))
+ image = Image.new("RGB", (800, 600), (225, 235, 250))
+ if kind != "covered":
+ painter.drawImage(ImageReader(image), 0, 0, 400, 300)
+ painter.saveState()
+ if kind == "transparent":
+ painter.setFillAlpha(0)
+ text = painter.beginText(20, 260)
+ text.setFont("Helvetica", 9)
+ text.setTextRenderMode(3 if kind == "hidden" else 0)
+ for line in [
+ "A real document may have a large decorative background image.",
+ "A searchable scan must still use the OCR inference pipeline.",
+ "Existing invisible text must not be adopted as native body text.",
+ ]:
+ text.textLine(line)
+ painter.drawText(text)
+ painter.restoreState()
+ if kind == "covered":
+ painter.drawImage(ImageReader(image), 0, 0, 400, 300)
+ painter.save()
+ image.close()
+ return stream.getvalue()
+
+
+ "kind,expected", [("background", "txt"), ("hidden", "ocr"), ("transparent", "ocr"), ("covered", "ocr")]
+)
[email protected]("effort", ["flash", "medium"])
+def test_real_auto_classification_controls_mineru_model_routing(
+ monkeypatch: pytest.MonkeyPatch, kind: str, expected: str, effort: str
+) -> None:
+ """只替换推理模型加载,真实 PDF 和 classify 决定解析模式及 Flash 原生分支。"""
+ calls = []
+
+ def model() -> SimpleNamespace:
+ """记录宿主推理初始化,背景 Flash 页应完全跳过该调用。"""
+ calls.append("model")
+ return SimpleNamespace(device="cpu")
+
+ monkeypatch.setattr(hybrid_runtime, "HybridLocalModelContextSingleton", lambda: SimpleNamespace(get_model=model))
+ monkeypatch.setattr(pipeline, "acquire_document", lambda *_: None)
+ monkeypatch.setattr(pipeline, "release_document", lambda *_: None)
+ monkeypatch.setattr(pipeline, "trim_process_heap", lambda: None)
+ state = pipeline._PDFAnalysis()
+ try:
+ pipeline._prepare_analysis(state, _image_layer_pdf(kind), effort, "auto", None)
+ assert state.parse_mode == expected
+ assert state.flash_txt_mode is (effort == "flash" and expected == "txt")
+ assert bool(calls) is (not state.flash_txt_mode)
+ finally:
+ pipeline._close_analysis(state)